Tuesday, November 12, 2013

NFL post-season correlations between wins and various stats

++Addition++Steve Sailer comments:
Regarding turnovers, another way to look at them is to separate interceptions and fumbles, and then look at total fumbles versus fumbles lost. It seems pretty likely that once the ball goes on the ground, it's sheer luck who winds up with it. (Both teams grapple like mad to get the ball.) So, the number of fumbles in a game is a measure of skill, but the percentage of fumbles lost is the most obvious and perhaps most important single measure of luck.
My emphasis on what strikes me as one hell of an insight into the game.

---

Having looked at several years of regular season NFL data, my working assumption is that, more than anything else, winning is about having an offense that is capable of moving the ball through the air, and that such an observation has not just contemporary relevance, but is historically descriptive as well (though the trend has become increasingly pronounced over time). It's boorish to so audaciously insinuating that I've discovered the key to success (not to mention a little painful, as I grew up as a votary of "Marty Ball"). If utilizing an occasional run for no other reason than to keep intact the integrity of the play action is how to become this year's Kansas City Chiefs, well--wait, at 9-0, they're 27th in passing and 12th in rushing. Like I said, boorish.

Speaking of our hometown Chiefs, juxtaposing the ball-controlling, defensively-led, conservative-play-calling Chiefs' style with the high-flying, shoot-out-welcoming Broncos is timely enough. I'd be happy to have this post thrown in my face this Sunday night, but if I had to bet on it, I'm not betting against Peyton Manning, aging agitated ankles be damned.

Sports commentators--many of them former players--regularly espouse phrases like "defenses win championships" and "to win, they'll need to maintain the ground game" in the post-season. Even if being able to pick apart the bottom of the division during the regular season gets a team into the playoffs, perhaps the dynamics are different at that elite-of-the-elite level.

Looking at the results from the last five years of post-season play (60 teams), the following stats correlate with playoff win percentage as relayed in the table below. Italics indicate a correlation in the 'unexpected' direction (average yards gained per rush play correlating inversely with winning, for example--that is, the more efficient a team has been moving the ball on the ground, the worse they tend to fare). Asterisks indicate a lack of statistical significance in the correlation at 95% confidence:

OffenseWins
Turnover ratio.62
Passer rating.52
Yards per pass play.45
Time of possession.31
Yards per play.27
3rd down conversion %.26
Total yards gained per game.25
Rush yards per game.24*
Pass yards per game.15*
1st downs per game.10*
Yards per rush play.10*
Defense
Rush yards allowed per game.42
3rd down conversion % allowed.34
Opposing passer rating.31
Yards allowed per pass play.28
Yards allowed per play.20*
Yards allowed per game.18*
Yards allowed per rush play.08*
Pass yards allowed per game.07*
1st downs allowed per game.02*

A few notes:

- Turnover ratio and time of possession are no more justifiably classified as offensive than as defensive.

- This methodological approach probably doesn't work as well in the post-season as it does during the regular season due to issues like home field advantage, #1 and #2 seeds having a buy week, more weather variation than during the regular season, etc, but it's still useful.

- Special teams stats are not included. They are far less informative than defensive and especially offensive numbers during the regular season. I expect nothing different in the post-season based on the similarities between offensive and defensive correlations during the regular seasons and post-seasons. Due to the time required to do this sort of data, I elected not to include special teams stats.

Relative to what we find during the regular season, turnovers are even more crucial in post-season play than they are during weeks 1-17. That's comprehensible given that teams are generally more evenly-matched during the playoffs than they are during the regular season, so giveaways are harder to overcome than they are during the regular season.

The quarterback spot is football's marquee position, and justifiably so. Just as is the case during the regular season, there is little that is as predictive of winning as a QB's passer rating in the playoffs. Using the bemusing formula has an advantage over other stats that are based primarily on simple averages--it negates the problems this sort of analysis has when teams intentionally run out the clock by keeping the ball on the ground, a phenomenon which tends to make winning teams' rush stats look relatively poor and losing teams' rushing defense looking correspondingly good, since both sides know what's coming.

The advantage yards-per-pass-play has over yards-per-rush-play isn't quite as pronounced as it is during the regular season, but it's still enormous.

Defensive stats simply aren't as explanatory as offensive stats are. Rush yards per game is often an artifice of the occurrence mentioned above--teams that give up a lot of yards on the ground tend to lose because winning teams tend to run a lot towards the end of the game against teams they're beating, even though said losing teams aren't necessarily any less efficient at stopping the run than winning teams are, as the weak relationship between winning and yards given up per rush play illustrates.

Opposing passer rating and yards per pass play, along with 3rd down conversion % allowed, are the only other major statistical indices considered that reach statistical significance, an expected converse of the success having an offense that is able to move the ball through the air brings.

4 comments:

Dan said...

Nice! As for the inverse correlations:

- Rush yards per play would be reduced by taking a knee in the victory formation repeatedly at the end of the game. Also, with a comfy lead, you will run the ball over and over at the end, being predictable to the opposing defense.

- Pass yards allowed can be high for winning teams. RGIII has had some great fantasy stats in losses, where he racks up yardage because his team is behind and he has to keep trying a comback. If you are forcing your opponent into the air, it means you are probably in the lead.

In both cases, winning *causes* those statistics late in the game.

Audacious Epigone said...

Dan,

Good point regarding the slop yards that losing teams are allowed to pick up through the air by prevent-style defenses, which not only allow for higher total pass yards to accumulate, but also probably inflate the losing QB's rating (though it does seem to me like a lot of picks get thrown by a team that is down by a couple of touchdowns and really airing it out with a couple of minutes remaining).

Steve Sailer said...

On the yards per rush being inversely correlated with winning, that may have something to do with the effectiveness of draw plays and the like against prevent defenses. The team in the lead is trying to avoid giving up a touchdown pass on defense and is happy to let the trailing team run the ball to keep the clock moving, so it's pretty easy to pick up 5 to 8 yards per running play when you really need to be passing.

Steve Sailer said...

Regarding turnovers, another way to look at them is to separate interceptions and fumbles, and then look at total fumbles versus fumbles lost. It seems pretty likely that once the ball goes on the ground, it's sheer luck who winds up with it. (Both teams grapple like mad to get the ball.) So, the number of fumbles in a game is a measure of skill, but the percentage of fumbles lost is the most obvious and perhaps most important single measure of luck.